Seasonal Fire Forecasting

One notable aspect of the MC1 Dynamic General Vegetation Model (DGVM) is the process-based fire module which simulates fire events and their impact on vegetation through time at regional to global scales. The module was built to explore the response of fire and its impacts to century-long scenarios of potential climatic change.

To project fire disturbance years into the future, the MC1 fire module was designed to dynamically simulate fuel availability, fuel moisture, and changes in vegetation type that constrain the spatial extent of fire. Fuel loading in different fuel classes is estimated using allometric functions of the live and dead biomass dynamically simulated by the MC1 biogeochemical module. The moisture content of each dead fuel class is simulated dynamically as a function of the climatic inputs to the model using standard equations that account for lags in dead fuel moisture according to fuel particle size. Live fuel moisture is modeled dynamically as a function of changes in soil water availability simulated by the MC1 biogeochemical module.

Potential fire behavior is estimated at each time step as a function of fuel loading and moisture using standard algorithms that calculate rate of fire spread and fire line intensity among other indices of fire behavior. Fire events are triggered by thresholds of potential fire behavior and an index of climatic drought. Fire effects on vegetation structure and biomass pools are estimated as a function of the simulated fire behavior and prorated as a function of the simulated spatial extent of a fire event. The spatial extent of a fire event (i.e., the burned proportion of a grid cell) is estimated as a function of time since the last fire event and by reference to minimum and maximum fire return intervals assigned to the current vegetation type as determined by the MC1 biogeography module.  

In 2004, the author of the MC1 fire module (J. Lenihan, USFS) received funding from the National Fire Plan to apply MC1 to the problem of seasonal-length fire forecasting for the conterminous USA. An MC1 fire forecasting system was designed in which observed monthly climate data are interpolated by the PRISM model to a relatively fine resolution (initially 50 km but currently 4 km resolution) modeling grid. These observed data grids are continuously updated at monthly intervals to incorporate newly available observations. Four to six 7-month weather forecasts are also obtained monthly from the web-based IRI data portal, and these relatively coarse-scaled forecast data are downscaled to the finer-scale modeling grid using an anomaly approach and a 30-year observed climatology. MC1 is run with the climatic data up to the last observed month. The results are then used to initiate MC1 runs for the 7-month period of each of the four to six weather forecasts. Consensus forecasts for fire-related variables are constructed from the combined results of individual forecast runs.

FIGURE LEGEND: Number of 7-month weather forecasts (out of five total) which result in high fire potential for the contiguous U.S. (blue=1, yellow=2, mustard=3, red=4, dark red=5)

Finally, the consensus forecast data are uploaded to Data Basin where they are graphically displayed as national-scale maps.  Datasets can be accessed directly in the Data Basin gallery "Fire, drought, and precipitation forecasting for the USA."

In addition, early season MC1 fire forecasts are routinely incorporated into the National Interagency Coordination Center’s assessments seasonal significant fire potential, and output from the forecasting system is also being used by other researchers for estimating fire suppression expenditures and optimal timing for prescribed fire (e.g. Program for Climate, Ecosystem and Fire Applications).

After a few seasons of comparing MC1 backcasts to annual observations of fire occurrence and extent in the conterminous USA, it was evident that the accuracy of the forecasts could be improved by assuming certain dynamic properties of the MC1 fire module were static for the purpose of seasonal-length forecasting. A problematical aspect of the fully dynamic mode is the allometric estimation of fuel bed loading and structure from the few coarsely-defined biomass pools simulated by the MC1 biogeochemical module. For example, standard fire behavior calculations used in the model are very sensitive to the “packing ratio”, a function of the mass vs. depth of the fuel bed, and yet there is little basis for estimating fuel bed depth from information internal to the model.

The fire behavior calculations are also sensitive to the total load of available fuel and it’s proportional distribution across six dead and live fuel classes. In the fully dynamic mode, these and other fuel bed related quantities are roughly estimated from biomass using values, limits, and proportional factors gleaned from standard fuel models and assigned to vegetation types simulated by the MC1 biogeography module. In the version of the MC1 fire module used for seasonal-length fire forecasting, the Wildland Fire Assessment System map of standard fuel model types and an associated lookup table are read by MC1. The standard values for fuel bed properties are used for estimating fire behavior and remain static, with one important exception. Annual fluctuations in total grass fuel load and seasonal transitions from live to dead fuel are simulated as a function of grass NPP and senescence dynamically-simulated by the MC1 biogeochemistry module. Values for tree height, crown length, bark thickness, and minimum and maximum fire return intervals used to calculate fire effects in MC1 are also considered static for the purposes seasonal-length fire forecasting, and their values are read from a Fuel Characteristics Classication System fuel bed map and an associated data lookup table.

For more information contact Dr. James Lenihan, USFS PNW.